search

xiph / flac

Free Lossless Audio Codec
https://xiph.org/flac/
GNU Free Documentation License v1.3
1.58k stars 278 forks source link

Further improve calculation of when to use wide residual computation #702

Closed ktmf01 closed 2 months ago

ktmf01 commented 2 months ago

This takes #700 one step further

In the following dataset, ./current is a binary without this PR and #700, ./mulbits is without this PR but with #700 and ./betterbps is including this PR. Testdata is the same as for #700: an upsampled set of short samples. Results are most pronounced for -5p on 24-bit material, see emphasis.

Command Mean [s] Min [s] Max [s] Relative
./betterbps -5 -j2 -c ../Rarewares-16bit.flac 1.214 ± 0.014 1.197 1.238 1.00
./mulbits -5 -j2 -c ../Rarewares-16bit.flac 1.223 ± 0.010 1.207 1.237 1.01 ± 0.01
./current -5 -j2 -c ../Rarewares-16bit.flac 1.217 ± 0.024 1.177 1.256 1.00 ± 0.02
./betterbps -5p -j2 -c ../Rarewares-16bit.flac 1.504 ± 0.035 1.463 1.552 1.24 ± 0.03
./mulbits -5p -j2 -c ../Rarewares-16bit.flac 1.495 ± 0.030 1.457 1.538 1.23 ± 0.03
./current -5p -j2 -c ../Rarewares-16bit.flac 1.497 ± 0.033 1.457 1.557 1.23 ± 0.03
./betterbps -8 -j2 -c ../Rarewares-16bit.flac 2.025 ± 0.014 2.005 2.050 1.67 ± 0.02
./mulbits -8 -j2 -c ../Rarewares-16bit.flac 2.002 ± 0.015 1.979 2.023 1.65 ± 0.02
./current -8 -j2 -c ../Rarewares-16bit.flac 2.011 ± 0.033 1.961 2.056 1.66 ± 0.03
./betterbps -8p -j2 -c ../Rarewares-16bit.flac 6.172 ± 0.050 6.107 6.285 5.08 ± 0.07
./mulbits -8p -j2 -c ../Rarewares-16bit.flac 6.176 ± 0.098 6.087 6.356 5.09 ± 0.10
./current -8p -j2 -c ../Rarewares-16bit.flac 6.297 ± 0.076 6.194 6.401 5.19 ± 0.09
./betterbps -5 -j2 -c ../Rarewares-24bit.flac 2.414 ± 0.042 2.356 2.476 1.99 ± 0.04
./mulbits -5 -j2 -c ../Rarewares-24bit.flac 2.637 ± 0.036 2.574 2.685 2.17 ± 0.04
./current -5 -j2 -c ../Rarewares-24bit.flac 2.851 ± 0.055 2.766 2.921 2.35 ± 0.05
./betterbps -5p -j2 -c ../Rarewares-24bit.flac 4.289 ± 0.059 4.188 4.388 3.53 ± 0.06
./mulbits -5p -j2 -c ../Rarewares-24bit.flac 6.038 ± 0.114 5.936 6.308 4.97 ± 0.11
./current -5p -j2 -c ../Rarewares-24bit.flac 8.807 ± 0.098 8.678 8.992 7.25 ± 0.12
./betterbps -8 -j2 -c ../Rarewares-24bit.flac 10.126 ± 0.043 10.051 10.170 8.34 ± 0.10
./mulbits -8 -j2 -c ../Rarewares-24bit.flac 10.556 ± 0.076 10.481 10.718 8.69 ± 0.12
./current -8 -j2 -c ../Rarewares-24bit.flac 10.654 ± 0.044 10.623 10.762 8.77 ± 0.11
./betterbps -8p -j2 -c ../Rarewares-24bit.flac 66.274 ± 0.240 65.983 66.583 54.58 ± 0.66
./mulbits -8p -j2 -c ../Rarewares-24bit.flac 74.999 ± 0.077 74.884 75.150 61.76 ± 0.72
./current -8p -j2 -c ../Rarewares-24bit.flac 81.717 ± 0.122 81.525 81.929 67.30 ± 0.79
Quake4 commented 1 month ago

Pls, fix max value by bits count to: FLAC__uint64 max_abs_sample_value = (((FLAC__uint64)1) << subframe_bps) - 1; Without types: (1 << subframe_bps) - 1